# HuggingNews HuggingNews is a news wire for AI and open source, where **the pipeline itself writes the stories**. Its own summary: *"Headlines and news articles, written from primary sources."* The site framing is simpler still: what is happening in AI today? ## How it works Three steps, and the middle one is the unusual part. **Collect.** It follows X accounts, labs, companies, projects, reporters and researchers, gathering posts and looking for clusters that might indicate a release, a product change, a funding announcement. **Detect by convergence.** A story is not declared because one outlet published it. It's declared when *multiple independent sources start writing about the same new thing*. Novelty plus corroboration, mechanically applied. **Write.** When sources converge on a genuinely new event, the pipeline researches it and produces a short article with a self-contained, scannable headline, grounded in the collected material and favouring primary sources over aggregators. Source material, a pull quote and a link to the original are shown alongside. Stories carry topics, tags, entities and sources, and the feed filters by topic, company, lab, project, person or product. Coverage spans models, research, open source, chips, infrastructure, agents, funding, policy, security and industry. ## Details worth noticing Two things visible in the page data that the marketing copy doesn't mention: **Stories record how many sources they came from.** A headline carries something like *"105 posts from 56 accounts"*. That number is a confidence signal you rarely get anywhere else, and it's exactly the thing you want when deciding whether something is real or one person's excitement. **Stories supersede each other.** Each entry carries flags for fresh, updated and superseded, with a pointer to the story it replaces. So the feed maintains a chain as a story develops rather than leaving you five near-duplicate headlines to reconcile. That's the hard part of news aggregation, and it's handled structurally. The copy is written by a fine-tuned Gemma model rather than a frontier API model, which explains how it stays economical at this volume. ## The agent angle There's a `SKILL.md` at `huggingnews.com/SKILL.md` you install with your agent, plus an optional API key (`HUGGINGNEWS_API_KEY`) that unlocks the full feed. So the wire is consumable two ways: read it yourself, or let an agent query it. Which is the right shape. A news source that only renders HTML is a source you have to visit; one that ships a skill is one your assistant can check on your behalf. ## Why I keep this note **The convergence heuristic is the transferable idea.** Wait until several independent sources are writing about the same new thing, then treat it as news. That is a genuinely good filter for an area that produces more announcements per week than anyone can read, and it's a rule I could apply to my own inputs without any tooling at all. **It's a working answer to a problem I have.** Keeping current on AI means either following hundreds of accounts or trusting a gatekeeper. This is a third option: follow the sources mechanically, surface what converges, cite everything. For newsletter and content work, a feed with source counts attached is a better starting point than a timeline. **And it's a live example of AI-written media done with some care.** Primary sources favoured, quotes shown, links back, an explicit note on judgment about covering concrete developments over commentary. The disclosure is upfront rather than buried. Whether the writing is *good* is a separate question worth checking against a story I already know well, which is the obvious test before relying on it. The caveat that follows: model-written summaries of model announcements is a short loop with no human in it. Treat it as a discovery surface pointing at primary sources, not as the source itself. The links are right there, which is the point. ## References - [huggingnews.com](https://huggingnews.com) — the feed - [About](https://huggingnews.com/about) — how the pipeline works - [SKILL.md](https://huggingnews.com/SKILL.md) — the agent skill ## Related - [[AI Agent Skills]] - [[AI Agents]] - [[Readwise]]